Hiroyuki Sato
Graduate School of Informatics and Engineering, The University of Electro-Communications, Japan
Minami Miyakawa
Faculty of Computer and Information Sciences, Hosei University (JSPS Research Fellow), Japan
Keiki Takadama
Graduate School of Informatics and Engineering, The University of Electro-Communications, Japan
Download article
http://dx.doi.org/10.3384/ecp171421074Published in: Proceedings of The 9th EUROSIM Congress on Modelling and Simulation, EUROSIM 2016, The 57th SIMS Conference on Simulation and Modelling SIMS 2016
Linköping Electronic Conference Proceedings 142:158, p. 1074-1080
Published: 2018-12-19
ISBN: 978-91-7685-399-3
ISSN: 1650-3686 (print), 1650-3740 (online)
For solving multi-objective problems, MOEA/D employs a set of weight vectors determining search directions and assigns one solution for each weight vector. Since the conventional MOEA/D assigns a randomly generated initial solution for each weight vector without considering its position in the objective space, mismatched pairs of initial solution and weight are generated, and it causes inef?cient search. To enhance MOEA/D based multi-objective optimization, this work proposes a method arranging randomly generated initial solutions to weight vectors based on positions of their solutions in the objective space. The proposed method is combined with the conventional MOEA/D and MOEA/D-CRU, and their search performances are veri?ed on continuous DLTZ4 benchmark problems with 2-5 objectives and different problem dif?culty parameters. The experimental results show that the proposed method improves the search performances of MOEA/D and MOEA/D-CRU especially on problems with the dif?culty to obtain uniformly distributed solutions in the objective space.
multi-objective optimization, many-objective optimization, evolutionary algorithm, MOEA/D
